'3LRN - Initial learning of S-R links: Instruction vs. Experience'
(AsPredicted #71133)
Download PDF

---

---

1) Have any data been collected for this study already?
It's complicated. We have already collected some data but explain in Question 8 why readers may consider this a valid pre-registration nevertheless.

2) What's the main question being asked or hypothesis being tested in this study?
How do representations of novel rule identities as described by novel stimulus-response, S-R associations, across different learning acquisition modes dynamically change in the brain? How is functional brain connectivity affected?
We aim at clarifying the nature of instruction-based learning behaviors, and at disentangling it from experience-based, trial-and-error learning. We hypothesize major patterns of activity in prefrontal and striatal regions for both instruction-based and trial-and-error learning. One caveat, however, resides in the fact that erroneously, previously tried responses are processed in the trial-and-error but not in the instruction-based learning condition. The third condition of interest, which is based on learning via passive observation of correct/incorrect S-R links, will possibly shed light on the way the implementation of incorrect responses might drive neural activity.

3) Describe the key dependent variable(s) specifying how they will be measured.
At the behavioral level, accuracy will be measured per each stimulus repetition level and learning condition. Accuracy is defined as 0 for incorrect and 1 for correct responses. In the instruction- and observation-based learning conditions, accuracy will be tracked only in the test phase.
As for the brain measures, the dependent variable is brain activity and connectivity per each learning condition and stimulus repetition. Brain activity will be acquired via whole-brain fMRI and tracked via univariate analysis, time-resolved MVPA, and functional connectivity analysis using beta-series correlations.

4) How many and which conditions will participants be assigned to?
There are 3 learning conditions across 24 experimental blocks. Each block is constituted by 4 different stimuli that are repeated 8 times across a learning phase (stimulus repetition 1 to 4) and a test phase (stimulus repetition 5 to 8).
The learning phase is what distinguishes one learning condition from the other in the way participants are asked to learn newly presented S-R associations. Each one of the 4 stimuli is associated with one unique button press that participants can learn via trial-and-error, instruction of the correct response, or observation of the correct/incorrect associations. The trial-and-error and observation-based learning conditions include trial-by-trial feedback during the learning phase about subjects' response accuracy and observed response, respectively.
The test phase does not include feedback and is consistent across learning conditions. Participants attend the same stimuli as in the learning phase and are asked to implement via button press the S-R associations they just learned.
Each participant will undergo all blocks in a randomized fashion across 3 functional MR runs.

5) Specify exactly which analyses you will conduct to examine the main question/hypothesis.
Behavioral performance as described by response accuracy will be analyzed via a mixed model/repeated measure, rmANOVA. Each model will include the independent variables stimulus repetition (levels 1 to 4 for learning and test phase) and learning condition.
Functional data will be slice-time corrected, spatially realigned, unwarped via field maps, and spatially normalized. The single-subject images will be smoothed before group analysis. The unsmoothed images will be used as input for the MVPA.
Univariate analysis: correct and incorrect trials will be modeled per each functional run, learning/test phase, condition, and stimulus repetition. First trials of each block will be modeled as separate events per each run. All trials in the learning phase of the instruction-based learning condition will be modeled as correct trials. Incorrect trials will not be modeled for the learning phase of this condition. In the observation-based learning condition, correct/incorrect trials are defined based on the observed response accuracy.
Brain activity statistical maps will be computed at individual and group levels in the context of the GLM. Via a rmANOVA, we will test for significant effects of learning condition and stimulus repetition at the whole-brain level.
Time-resolved MVPA: for each individual statistical map, condition-specific single-trial BOLD activity will be estimated via the least-square-separate, LSS approach. We will estimate as many LSS models as the number of trials. The voxel-wise beta estimates for each stimulus repetition level will be compared via rmANOVA: within each learning phase (stimulus repetition 1 & 2 vs 3 & 4 in the learning phase, and stimulus repetition 5 & 6 vs 7 & 8 in the test phase), and between phases, by looking at the transition from the learning to the test phase (stimulus repetition 3 & 4 of the learning phase vs 5 & 6 of the test phase). T-tests will explore directed hypotheses about the ANOVA results. For example, we are interested in testing whether the expected significant effects for region and stimulus repetition could exhibit a decreasing activity trend across stimulus repetition levels in the lateral PFC or an increase in the ventral striatum.
The MVPA will be computed within anatomically predefined regions of interest, ROIs, including the ventrolateral and dorsolateral PFC.
Connectivity analysis: we will compare the same single-trial beta estimates we used for the MVPA to look at changes in functional connectivity during learning and from the learning to the test phase (stimulus repetition 1 & 2 vs stimulus repetition 7 & 8). Single-trial betas will be compared between seed regions and between one seed region and the rest of the brain. We are also interested in tracking slower vs faster connectivity changes. To this aim, we will compute changes in correlation values between other repetition levels.

6) Describe exactly how outliers will be defined and handled, and your precise rule(s) for excluding observations.
Participants without a full dataset will be discarded.
For properly modeling the observation-based learning condition as described in Question 8, data will be discarded from participants who did not learn in the trial-and-error condition. This will be determined by a response pattern analysis of the trial-and-error blocks. The inclusion criteria will be set to 75% correct responses to the same stimulus across test and learning phases per each block. Blocks with 3 or more inaccurate responses for the same stimulus in the test phase, will be inspected in the respective learning phase. Here, a minimum 50% response accuracy for the same stimulus will determine whether data from this subject will be excluded. In the trial-and-error condition, one single "bad" block as described above will determine the fate of the full subject's dataset.
Blocks in the other conditions that have less than 75% accuracy in the test phase in at least 50% of the stimuli presented will be excluded.
Participants with individual statistical maps that do not show significant activity (p<.001 uncorrected) in visual and left motor cortices when inspected via T-contrasts testing for significant voxels across conditions will be discarded.

7) How many observations will be collected or what will determine sample size?
No need to justify decision, but be precise about exactly how the number will be determined.
Participants will be tested in blocks of 5. Data will be acquired until a number of 80 usable datasets is reached.

8) Anything else you would like to pre-register?
(e.g., secondary analyses, variables collected for exploratory purposes, unusual analyses planned?)
Data from 7 subjects have been already acquired. The first pilot subject will not be included in the dataset and served to ensure the proper design and logging.
Data from the other 6 participants will be included in the dataset. Response patterns in the trial-and-error condition from the first block of 5 participants were used to prepare the stimulus material for the observation-based learning condition for the next 5 (subject number 6 to 10). Stimulus material preparation for the observation-based learning condition will be an ongoing, highly controlled process for the entire duration of the data acquisition period. Data acquisition from these first participants served to estimate the time required for experimental material preparation and to test the feasibility of acquiring data in this fashion given the scanner booking constraints.
Explorative analyses: T-tests will be implemented to test the direction of the significant effects expected from the mixed model/ANOVA in the context of the behavioral performance analysis. For example, if a significant effect is found for stimulus repetition level, then we will implement T-tests to contrast one repetition level to another.
T-tests will be used to explore the direction of any significant effect that is apparent from the ANOVA in the context of the univariate analysis.
We will compute an exploratory whole-brain univariate analysis to possibly define additional ROIs for the MVPA and connectivity analysis.
A whole-brain searchlight will explore possible additional effects outside the ROIs.
In an explorative fashion and to possibly confirm the results of the MVPA output, we might collapse data across learning conditions and/or stimulus repetitions to increase power and find possible stronger effects.